P=NP
Article
P=NP is a recurring concept in the Astral Codex Ten archive, appearing 2 times across 2 issues between February 15, 2023 and July 26, 2025. The archive places it in contexts such as “I get emails every single day from P=NP crackpots”; “explaining why P=NP is a difficult problem”. It most often appears alongside ACX, ivermectin, 4chan.
Metadata
- Category: Concepts
- Mention count: 2
- Issue count: 2
- First seen: February 15, 2023
- Last seen: July 26, 2025
Appears In
Related Pages
-
- ACX (2 shared issues)
-
- ivermectin (2 shared issues)
-
- 4chan (1 shared issues)
-
- ACX (1 shared issues)
-
- ACX Commentariat (1 shared issues)
-
- ACX-era (1 shared issues)
-
- AI (1 shared issues)
-
- Alexander (1 shared issues)
-
- alt-right (1 shared issues)
-
- an alternative forum (1 shared issues)
-
- Anatoly (1 shared issues)
-
- Anthony Fauci (1 shared issues)
External Links
None.
Source Context
Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.
This is a great post that contains a lot of truth. And yet … I also see a grain of truth in Kavanagh’s position. Like, I get emails every single day from P=NP crackpots and quantum mechanics crackpots and now AI crackpots too. Some of them probably *would* be better off never trying to think for themselves again, and just Trusting Science and Trusting the Experts. Sure, the experts are sometimes confidently wrong, but not as consistently so as they are! And for my part, I can’t possibly write 25,000 words to explain why each and every crackpot is wrong. As a matter of survival, I *have* to adopt a Kavanagh-like heuristic: “this person seems like an idiot.”
I think, in hindsight, that this is reflecting a unique characteristic of the ACX Commentariat, which is that it is unusually likely to develop an idea or conceptual schema rather than just asserting something and moving on. For example, here’s a Comment of the Week where Anatoly spends a very long time explaining the different meanings of ‘infinite’ and ‘finite’ in the context of explaining why P=NP is a difficult problem. It has a Brunet Index of 14.3 (so a little less than double the local average) because it repeats the words ‘infinite’, ‘finite’ and ‘algorithm’ many times. But I agree with the responses that this is a great comment, and exactly the sort of thing which only the ACX Commentariat seems to produce with any regularity. For a more recent example, here’s another comment of the week by Benjamin Jolley which adds some details to Scott’s post The Compounding Loophole, and is also clearly a great post which fits very well into the Commentariat corpus. So my conclusion here is that documentation for these tests assumes that stale vocabulary is always bad, because it expects you to be using the tests on – for example - novels. However, stale vocabulary isn’t inherently good or bad, and in this case it serves as a marker for something the Commentariat like or value. Anecdotally, it looks like what the Commentariat value is something like ‘well defined terms’. Even if this doesn’t map cleanly to something we can point to, there’s no accounting for taste - if the Commentariat just happen to prefer lengthy stale sentences there’s nothing actually wrong with that. Therefore, this measure is consistent with the other measures of complexity even though it very clearly shows the opposite relationship than I expected. Just for fun, I thought I would show the most repetitive comment ever written. This was actually slightly difficult as there are a lot of things which are both comments and repetitive but which would be uninteresting to count (spam, code snippets, pasted text from early LLMs where the model hangs and repeats the same text to infinity). The most repetitive non-spam comment which I reckon was generated by humans alone is this comment by Deiseach, which quotes extensively from an early Irish law book (Brunet = 16.9). The most repetitive non-spam comment which I reckon has a single human author is this comment by Fahundo (Brunet = 16.5), giving the answer to a logic problem in ROT13 (so actually possibly breaks the rule about not using a computer in the writing, but not in the way I meant!) Complexity Approach 4 – Reading age Finally, I looked at reading age, although this approach was largely unsuccessful. ‘Reading age’ is an approximate composite measure of the complexity of language and sentence construction in a piece of text. There are quite a few different measures of reading age, which all show roughly the same outcome in my data. The one I have depicted below is the Flesh-Kincaid Grade level, which roughly tracks how many years of continuous schooling you would theoretically need to read and understand the text. The Commentariat is a largely very intellectual bunch and so a typical reading age of around 10.5 is unsurprising (a typical SSC/ACX comment is just barely less complex than an academic article in terms of vocabulary and construction, and the most complex comments significantly exceed this). The graph shows that comment complexity jumps by approximately half a grade level when SSC becomes ACX, but I’m a bit sceptical this is a ‘real’ effect. Most reading age formulae track sentence length very closely, and for some reason sentence length also changes significantly around this time. I could genuinely believe that sentence length changes on the switch to ACX, but I don’t think measures of reading age are designed to be valid if sentence length is changing for reasons unrelated to the complexity of text, so I don’t think you can confidently conclude the ACX comments are more sophisticated from this measure alone. Complexity - Conclusions Overall, it is appropriate to discover that my measure of ‘complexity of thought’ is itself complex. We do see very clear peaks in the SSC era, but not actually quite in the place we expected to see them. Similarly, we don’t always see the peak in the direction we expect (sentences are long and stale in the peak SSC years, which doesn’t seem like a recipe to promote engagement). Finally, we have a puzzle about how the Substack UI/UX causes significantly fewer sentences per comment. My conclusion here is that these data are completely consistent with a Commentariat who have a particular thing that they like, which peaked in 2017. This thing quantitively looks like long stale sentences, but actually might qualitatively feel different – like for example careful definitions of words which are then used repeatedly. As for why the peak sentence length is after peak engagement, my best guess is that people didn’t stop engaging at random; the people with the strongest commitment to the Commentariat stuck around longest, and these are also the people with the most respect for SSC cultural norms (leave long, thoughtful comments) and willingness to dedicate time to commenting. I have heard this described as ‘evaporative cooling’ before. This group of ‘fanatics’ hung around for a bit longer than everyone else, but eventually either they mostly left too or their influence on discussion norms was not strong enough to prevent a reversion towards the comment section mean (which tends towards shorter and less rigorous comments) What happened in 2016? From the data I extracted, it is clear something happened to the Commentariat in 2016(ish) and again in 2021 with the switch to ACX. Of the four measures I presented: Depth of engagement shows two clear directional reversals in 2016 and 2021